rformance, while avoiding 160 overfitting the model to any location 161 Materials and Methods 162 Data Preparation 163 We used data from the Genomes to Fields (G2F) initiative for years 2014-2019 164 (McFarland et al. 2020), focusing on the sites within the continental United States. Each year’s 165 data are publicly available (https://www.genomes2fields.org/resources/), including weather and 166 soil data for field sites, genomic data, management schedules (e.g., application of fertilizer, 167 herbicides, irrigation) and yield (in addition to other phenotypic variables). We augmented this 168 through additional genomic and weather data. Weather data retrieved from Daymet (Thornton et 169 al
Open resource ↗Genomes to Fields · pdf-raw-page:7 lines:1-53Unverified paper record
Yield Prediction Through Integration of Genetic, Environment, and Management Data Through Deep Learning
bioRxiv · 30 Jul 2022 · 10.1101/2022.07.29.502051
Abstract
Accurate prediction of the phenotypic outcomes produced by different combinations of genotypes, environments, and management interventions remains a key goal in biology with direct applications to agriculture, research, and conservation. The past decades have seen an expansion of new methods applied towards this goal. Here we predict maize yield using deep neural networks, compare the efficacy of two model development methods, and contextualize model performance using linear models, which are the conventional method for this task, and machine learning models We examine the usefulness of incorporating interactions between disparate data types. We find a deep learning model with interactions has the best average performance. Optimizing submodules for each datatype improved model performance relative to optimizing the whole model for all data types at once. Examining the effect of interactions in the best performing model revealed that including interactions altered the model’s sensitivity to weather and management features, including a reduction of the importance scores for timepoints expected to have limited physiological basis for influencing yield – those at the extreme end of the season, nearly 200 days post planting. Based on these results, deep learning provides a promising avenue for phenotypic prediction of complex traits in complex environments and a potential mechanism to better understand the influence of environmental and genetic factors.
Plant phenotyping relevance
遺伝子型・環境・管理データからトウモロコシ収量という植物形質を予測する深層学習手法を開発・比較しており、表現型推定が研究の中心である。
abstractHere we predict maize yield using deep neural networks, compare the efficacy of two model development methods, and contextualize model performance using linear models, which are the conventional method for this task, and machine learning models
abstractBased on these results, deep learning provides a promising avenue for phenotypic prediction of complex traits in complex environments
Code and data availability
The paper uses publicly available Genomes to Fields (G2F) maize phenotype/weather/soil data (2014-2019) and provides authors' custom Python processing/analysis scripts on two public Bitbucket repositories. A Zenodo deposit (10.5281/zenodo.6916775) with PCA eigenvectors is mentioned but its URL is not among the allowed,
the eigenvectors 400 resulting from the principal components analysis are provided to enable transformation of 401 provided genomes. Additional weather measurements were retrieved from Daymet (Thornton et 402 al. 2020). Custom python scripts for downloading, aggregating and processing these data are 403 available on bitbucket (https://bitbucket.org/washjake/maizemodel and 404 https://bitbucket.org/daniel_kick/maizemodel/ ) in the notebooks directory (files with the prefix 405 0.0 to 0.5). 406 407 and is also made available for use under a CC0 license. was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC 105
Open resource ↗washjake/maizemodel · pdf-raw-page:17 lines:1-46scripts were used to 173 aggregate and standardize terminology across years. Rather than itemizing each operation, we 174 restrict ourselves to those which are likely to be of interest to those working with similar data 175 sets. The scripts used are available through Bitbucket 176 (https://bitbucket.org/washjake/maizemodel and https://bitbucket.org/daniel_kick/maizemodel/ ). 177 Scripts were written in Python (Van Rossum and Drake 2009 p. 3) and rely on scientific and 178 common general libraries (Seabold and Perktold 2010; Pedregosa et al. 2011; fuzzywuzzy 179 2017; Virtanen et al. 2020; team 2020; Harris et al. 2020; Da Costa-Luis et al. 2022) along with 180 plotting libraries for explora
Open resource ↗daniel_kick/maizemodel · pdf-raw-page:8 lines:1-57This is an automatically classified, unverified record. Curator approval is required before any resource enters the Catalog.